Connection Draining (Deregistration Delay) in AWS Load Balancers

Overview

Connection Draining (for Classic Load Balancers) or Deregistration Delay (for Application Load Balancers and Network Load Balancers) is a feature that ensures in-flight requests are completed before an instance is removed from service. This mechanism prevents abrupt connection drops when an instance is deregistered or marked unhealthy.

How It Works

1. When an instance enters draining mode, the Elastic Load Balancer (ELB) stops routing new requests to it.
2. Existing connections are given time (draining period) to complete their active requests.
3. Once the draining period expires or all connections are closed, the instance is fully deregistered.
4. Any new incoming requests are routed to other available instances.

Example Scenario

• Imagine you have three EC2 instances behind an ELB.
• One instance is placed in draining mode.
• Existing users connected to that instance finish their requests within the draining period.
• The ELB ensures that new users are directed to the remaining healthy instances.

Configuring Connection Draining

You can configure the Connection Draining time between 1 and 3,600 seconds (5 minutes by default).

• Set to 0 → Disables draining, immediately closing all active connections.
• Low values (e.g., 30 seconds) → Best for short-lived requests (e.g., small API calls).
• High values (e.g., several minutes) → Suitable for long-running requests (e.g., file uploads, streaming).

Key Considerations

• Shorter draining times help remove instances quickly but may interrupt long requests.